Numerical Methods For Modeling Transient Flow: in Distribution Systems

104 JULY 2005 | JOURNAL AWWA 97: 7 | PEER-REVI EWED | WOOD ET AL
2005 American Water Works Association

Numerical methods
for modeling transient flow
in distribution systems
The authors compared the formulation and computational performance of two numerical
methods for modeling hydraulic transients in water distribution systems. One method is
Eulerian-based, and the other is Lagrangian-based. The Eulerian approach explicitly solves
the hyperbolic partial differential equations of continuity and momentum and updates the
hydraulic state of the system in fixed grid points as time is advanced in uniform increments.
The Lagrangian approach tracks the movement and transformation of pressure waves and
updates the hydraulic state of the system at fixed or variable time intervals at times when
a change actually occurs. Each method was encoded into an existing hydraulic simulation
model that gave initial pressure and flow distribution and was tested on networks of varying
size and complexity under equal accuracy tolerance. Results indicated that the accuracy
of the methods was comparable, but that the Lagrangian method was more computationally
efficient for analysis of large water distribution systems.
BY DON J. WOOD,
SRINIVASA LINGIREDDY,
PAUL F. BOULOS,
BRYAN W. KARNEY,
AND DAVID L. MCPHERSON
omputer models for simulating the hydraulic and water quality behav-
ior of water distribution systems have been available for many years.
Recently these models have been extended to analyze hydraulic tran-
sients as well. In the past, most transient analyses were performed only
on large transmission mains using highly skeletonized models for engi-
neering design; proprietary transient computer programs were largely limited to
specialist consulting companies, research organizations, and universities.
In addition to improved design and operation of water distribution systems,
a driving force behind the trend toward increased analysis of hydraulic tran-
sients has been the growing awareness that hydraulic transients can create unex-
pected opportunities for pathogens present in the external environment to intrude
into the distribution system with disastrous consequences to public health. Mod-
ern management of water distribution systems requires simulation models that
are able to accurately predict transient flow and pressure variations within the dis-
tribution system environment.
PRESSURE TRANSIENTS
Effect on water quality. It is well-recognized that pressure transients may
adversely affect the quality of treated water. Pressure transients in water distri-
bution systems result from an abrupt change in the flow velocity and can be
caused by main breaks, sudden changes in demand, uncontrolled pump starting
C
water storage and distribution
or stopping, fire hydrant opening and
closing, power failure, air-valve slam,
flushing operations, feed-tank drain-
ing, overhead storage tank loss, pipe
filling and draining, and other condi-
tions (Karim et al, 2003). These events
can generate high intensities of fluid
shear and may cause resuspension of
settled particles as well as biofilm
detachment. So-called red water events
have often been associated with tran-
sient disturbances.
Moreover, a low-pressure transient
eventarising from a power failure
or pipe break, for examplehas the
potential to cause contaminated
groundwater to intrude into a pipe
at a leaky joint or break. Depending
on the size of the leaks, the volume of
intrusion can range from a few gal-
lons to hundreds of gallons (Funk et
al, 1999; LeChevallier, 1999). Neg-
ative pressures induce backsiphonage
of nonpotable water from domestic,
industrial, and institutional piping
into the distribution system. Formation of vapor cavi-
ties during low-pressure transients and the subsequent
collapse of vapor cavities could result in cavitation
corrosion or damage to the pipes protective film
(USACE, 1999).
If not properly designed and maintained, even some
common transient-protection strategies (such as relief
valves or air chambers) may permit pathogens or other
contaminants to find a route into the potable water
distribution system. Similarly, increasing overhead stor-
age for surge protection (e.g., a closed tank, open stand-
pipe, feed tank, or bladder tank) can result in long res-
idence times, which in turn may contribute to water
quality deterioration. Deleterious effects include chlo-
WOOD ET AL | PEER-REVI EWED | 97: 7 JOURNAL AWWA | JULY 2005 105
Data
Valve
Orifice
H
R
H
O
q
O
H
ex
FIGURE 1 Pipeline and data for comparison of MOC, WCM, and
exact solution
H
ex
hydraulic grade at pipe exit, H
o
hydraulic grade outside the orifice, H
R
reservoir
hydraulic grade, MOCmethod of characteristics, q
o
initial flow rate through the
pipeline, WCMwave characteristic method
Case 1 (no orifice): H
R
= H
o
= 45 ft (13.7 m), H
ex
= 0; case 2 (orifice): H
R
= 135 ft (41.2
m), H
o
= 45 ft (13.7 m), H
ex
= 0; length = 3,600 ft (1,098 m), diameter = 12 in. (300 mm),
capacity = 3,600 fps (1,098 m/s), q
o
= 3 cfs (85 L/s).
Exact MOC WCM
Hydraulic Volumetric Pressure Volumetric Pressure Volumetric
Time Ratio Grade Flow Rate Head Flow Rate Head Flow Rate
s * ft (m) cfs (L/s) ft (m) cfs (L/s) ft (m) cfs (L/s)
0 1.00 45.00 (13.72) 3.00 (84.90) 45.00 (13.72) 3.00 (84.90) 45.00 (13.72) 3.00 (84.90)
1 0.84 59.13 (18.02) 2.90 (82.07) 59.13 (18.02) 2.90 (82.07) 59.13 (18.02) 2.90 (82.07)
2 0.69 79.60 (24.26) 2.76 (78.11) 79.60 (24.26) 2.76 (78.11) 79.60 (24.26) 2.76 (78.11)
3 0.55 98.62 (30.06) 2.42 (68.49) 98.62 (30.06) 2.42 (68.49) 98.62 (30.06) 2.42 (68.49)
4 0.41 117.91 (35.94) 2.00 (56.60) 117.91 (35.94) 2.00 (56.60) 117.91 (35.94) 2.00 (56.60)
5 0.29 126.69 (38.62) 1.47 (41.60) 126.69 (38.62) 1.47 (41.60) 126.69 (38.62) 1.47 (41.60)
6 0.19 122.48 (37.33) 0.95 (26.89) 122.48 (37.33) 0.95 (26.89) 122.48 (37.33) 0.95 (26.89)
7 0.11 102.82 (31.34) 0.49 (13.87) 102.82 (31.34) 0.49 (13.87) 102.82 (31.34) 0.49 (13.87)
8 0.05 75.08 (22.88) 0.19 (5.38) 75.08 (22.88) 0.19 (5.38) 75.08 (22.88) 0.19 (5.38)
9 0.10 51.89 (15.82) 0.04 (1.13) 51.89 (15.82) 0.04 (1.13) 51.89 (15.82) 0.04 (1.13)
10 0.00 41.92 (12.78) 0.00 (0.00) 41.92 (12.78) 0.00 (0.00) 41.92 (12.78) 0.00 (0.00)
MOCmethod of characteristics, WCMwave characteristic method
*ratio of the effective flow area to the fully open flow area for the butterfly valve
TABLE 1 Case 1: Single pipe leading from a reservoir to a valve
rine residual loss and possible increases in the concen-
tration of microorganisms (Clark et al, 1996). Pathogens
can also enter the distribution system during construc-
tion, repair, cross-connections, and conditions and activ-
ities in which the system is open to the atmosphere or
the environment (Karim et al, 2003). The effects of
pressure transients on distribution system water qual-
ity degradation have been extensively reviewed (Wood
et al, 2005; Boulos et al, 2004; Karim et al, 2003;
LeChevallier et al, 2003; Kirmeyer et al, 2001; Funk
et al, 1999).
MODELING APPROACHES
Approaches to modeling pressure
transients. Several approaches have
been taken to numerically model the
movement and transformation of pres-
sure waves in water distribution sys-
tems and can be classified as either
Eulerian or Lagrangian. The Eulerian
approach reformulates the governing
transient flow equations (discussed
subsequently) into total differential
equations, which are then expressed in
a finite difference form. The
Lagrangian approach tracks changes
in pressure waves (having both posi-
tive and negative amplitude) as they
travel through the pipe network and
updates the state of the network only
at times when a change actually
occurs, such as when a pressure wave
reaches the end node of a pipe.
The computational accuracy and performance of these
hydraulic transient modeling approaches have not been
comprehensively compared using an objective testing pro-
tocol applied to reasonably sized networks. Certainly,
well-known numerical challenges in solving the transient
flow equations exist, for example avoiding numerical dis-
persion and attenuation and eliminating unnecessary dis-
tortion of either the physical pipe system or its boundaries.
Yet even with the continuous advances in computing
speed and capacity, development of efficient network-
modeling methods has lagged for several reasons.
250
(76.20)
200
(60.96)
150
(45.72)
100
(30.48)
50
(15.24)
0
0 1 2 3 4 5 6 7 8 9 10
Times
Case 1
Case 2
FIGURE 2 Pressure head variations at valve
P
r
e
s
s
u
r
e

H
e
a
d
f
t

(
m
)
Exact MOC WCM
Hydraulic Volumetric Pressure Volumetric Pressure Volumetric
Time Ratio Grade Flow Rate Head Flow Rate Head Flow Rate
s * ft (m) cfs (L/s) ft (m) cfs (L/s) ft (m) cfs (L/s)
0 1.00 45.00 (13.72) 3.00 (84.90) 45.00 (13.72) 3.00 (84.90) 45.00 (13.72) 3.00 (84.90)
1 0.84 59.13 (18.02) 2.90 (82.07) 59.13 (18.02) 2.90 (82.07) 59.13 (18.02) 2.90 (82.07)
2 0.69 79.61 (24.27) 2.76 (78.11) 79.61 (24.27) 2.76 (78.11) 79.61 (24.27) 2.76 (78.11)
3 0.55 104.68 (31.91) 2.50 (70.75) 104.68 (31.91) 2.50 (70.75) 104.68 (31.91) 2.50 (70.75)
4 0.41 136.09 (41.48) 2.15 (60.85) 136.09 (41.48) 2.15 (60.85) 136.09 (41.48) 2.15 (60.85)
5 0.29 169.59 (51.69) 1.71 (48.39) 169.59 (51.69) 1.71 (48.39) 169.59 (51.69) 1.71 (48.39)
6 0.19 197.60 (60.23) 1.20 (33.86) 197.60 (60.23) 1.20 (33.86) 197.60 (60.23) 1.20 (33.86)
7 0.11 207.54 (63.26) 0.70 (19.81) 207.54 (63.26) 0.70 (19.81) 207.54 (63.26) 0.70 (19.81)
8 0.05 189.91 (57.88) 0.30 (8.49) 189.91 (57.88) 0.30 (8.49) 189.91 (57.88) 0.30 (8.49)
9 0.10 152.07 (46.35) 0.07 (1.98) 152.07 (46.35) 0.07 (1.98) 152.07 (46.35) 0.07 (1.98)
10 0.00 123.20 (37.55) 0.00 (0.00) 123.20 (37.55) 0.00 (0.00) 123.20 (37.55) 0.00 (0.00)
*ratio of the effective flow area to the fully open flow area for the butterfly valve
TABLE 2 Case 2: Single pipe leading from a reservoir with an orifice to a valve
Model criteria. First, the trend to
include dead ends for transient con-
sideration and to develop all pipe mod-
els for improved water quality charac-
terization increases the size of the
model to be solved. Similarly, the need
to directly interface hydraulic network
models with computer-assisted design
and geographical information system
applications (e.g., work-order and
maintenance-management systems)
requires compatibility with large data
sets and the evaluation of an all-pipe
model. Finally, the trend toward link-
ing real-time supervisory control and
data acquisition systems, data loggers,
and graphical user interfaces to net-
work models with interactive analysis
and graphical display of results
demands that solutions to network
models be obtained as quickly and
accurately as possible.
This research reviewed and com-
pared the two types of transient models
for pipe networks, i.e., Eulerian-based
and Lagrangian-based. The models
were contrasted with respect to how
closely their results matched analytical
solutions, how closely the results
matched one another, how long the
models took to execute, and how many
calculations were required.
EULERIAN AND LAGRANGIAN
APPROACHES
TO TRANSIENT FLOW ANALYSIS
Governing equations. Rapidly vary-
ing pressure and flow conditions in
pipe networks are characterized by
variations that are dependent on both
position, x, and time, t. These condi-
tions are described by the continuity
equation
H
t
=

g
c
A
2
Q
x
(1)
and the momentum (Newtons second
law) equation
H
x
=

g
1
A
Q
t

+ f(Q) (2)
in which H is the pressure head (pres-
sure/specific weight), Qis the volumet-
ric flow rate, c is the sonic wave speed
in the pipe, A is the cross-sectional area,
1
2
4 4
5 9
6
8
7
5 2
3
3
Valve
6
626.64 ft (191 m)
Reservoir
FIGURE 3 Pipe network schematic and pressure head at node 3
and valve for Example 2
A
Pipe number
Node number
0 5 10 15 20 25 30 35 40 45 50
Times
Lagrangian WCM Eulerian MOC
H
e
a
d
f
t

(
m
)
0
200
400
600
800
1,000
1,200
B
Node 3
(365.76)
(304.80)
(243.84)
(182.88)
(121.92)
(76.20)
0 5 10 15 20 25 30 35 40 45 50
Times
P
r
e
s
s
u
r
e

H
e
a
d
f
t

(
m
)
0
200
(60.96)
400
(121.92)
600
(182.88)
800
(243.84)
1,000
(304.80)
1,200
(365.76)
1,400
(426.72)
C
At valve
Eulerian MOC Lagrangian WCM
Valve closure = 0.6 s
g is the gravitational acceleration, and f(Q) represents a
pipe-resistance term that is a nonlinear function of flow
rate. Eqs 1 and 2 have been simplified by considering
changes only along the pipe axis (one-dimensional flow)
and discarding terms that can be shown to be of minor sig-
nificance. A transient flow solution is obtained by solv-
ing Eqs 1 and 2 along with the appropriate initial and
boundary conditions. However, except for very simple
applications that neglect or greatly simplify the boundary
conditions and the pipe-resistance term, it is not possible
to obtain a direct solution. When pipe junctions, pumps,
surge tanks, air vessels, and other components that rou-
tinely need to be considered are included, the basic equa-
tions are further complicated, and it is necessary to utilize
numerical techniques. Accurate transient analysis of large
pipe networks requires computationally efficient and
accurate solution techniques.
Both Eulerian and Lagrangian solution schemes are
commonly used to approximate the solution of the gov-
erning equations. Eulerian methods update the hydraulic
state of the system in fixed grid points as time is advanced
in uniform increments. Lagrangian methods update the
hydraulic state of the system at fixed or variable time
intervals at times when a change actually occurs. Each
approach assumes that a steady-state hydraulic equilib-
rium solution is available that gives initial flow and pres-
sure distributions throughout the system.
Eulerian approach. Eulerian methods consist of the
explicit method of characteristics (MOC), explicit and
implicit finite difference techniques, and finite element
methods. For pipe network (closed
conduit) applications, the most well-
known and widely used of these tech-
niques is the MOC (Boulos et al,
2004). The MOC is considered the
most accurate of the Eulerian meth-
ods in its representation of the gov-
erning equations but requires numer-
ous steps or calculations to solve a
typical transient pipe-flow problem.
As the pipe system becomes more
complex, the number of required cal-
culations increases, and a computer
program is required for practical
applications. This method has been
summarized by other researchers
(Boulos et al, 2004; Larock et al,
1999; Chaudhry, 1987; Watters,
1984; Streeter & Wylie, 1967) and implemented in var-
ious computer programs for pipe system transient analy-
sis (Axworthy et al, 1999; Karney & McInnis, 1990).
Lagrangian approach. The Lagrangian approach solves
the transient flow equations in an event-oriented system-
simulation environment. In this environment, the pressure
wave propagation process is driven by the distribution
system activities. The wave characteristic method (WCM)
is an example of such an approach (Wood et al, 2005;
Boulos et al, 2004) and was first described in the litera-
ture as the wave plan method (Wood et al, 1966). The
method tracks the movement and transformation of pres-
sure waves as they propagate throughout the system and
computes new conditions either at fixed time intervals
or at times when a change actually occurs (variable time
intervals). The effect of line friction on a pressure wave is
accounted for by modifying the pressure wave using a
nonlinear characteristic relationship describing the cor-
responding pressure head change as a function of the
lines flow rate. Although it is true that some approxi-
mation errors will be introduced using this approach,
these errors can be minimized using a distributed-fric-
tion profile (piecewise linearized scheme).
However, this approach normally requires orders of
magnitude fewer pressure and flow calculations, which
allows very large systems to be solved in an expeditious
manner, and has the additional advantage of using a sim-
ple physical model as the basis for its development.
Because the WCM is continuous in both time and space,
the method is also less sensitive to the structure of the
Pipe Length Diameter
Number ft (m) in. (mm) Roughness Minor Loss
1 2,000 (610) 36 (900) 92 0
2 3,000 (914) 30 (750) 107 0
3 2,000 (610) 24 (600) 98 0
4 1,500 (457) 18 (450) 105 0
5 1,800 (549) 18 (450) 100 0
6 2,200 (671) 30 (750) 93 0
7 2,000 (610) 36 (900) 105 0
8 1,500 (457) 24 (600) 105 0
9 1,600 (488) 18 (450) 140 0
TABLE 3 Pipe characteristics for example 2
Modern management of water distribution systems
requires simulation models that are able to accurately predict
transient flow and pressure variations within the distribution
system environment.
network and to the length of the simulation process,
resulting in improved computational efficiency. This tech-
nique produces solutions for a simple pipe system that are
virtually identical to those obtained from exact solutions
(Boulos et al, 1990). A similar comparison of exact and
numerical results is presented in this article.
MOC strategy. In the strategy used by the MOC, the
governing partial differential equations are converted to
ordinary differential equations and then to a different
form for solution by a numerical method. The equations
express the head and flow for small time steps (t) at
numerous locations along the pipe sections. Calculations
during the transient analysis must begin with a known ini-
tial steady state and boundary conditions. In other words,
head and flow at time t = 0 will be known along with head
and/or flows at the boundaries at all times. To handle
the wave characteristics of the transient flow, head and
flow values at time t + t at interior locations are calcu-
lated making use of known values of head and flow at the
previous time step at adjacent locations using the ordinary
differential equations expressed in different form.
Calculating WCM concept. The WCM is based on the
concept that transient pipe flow results from the genera-
tion and propagation of pressure waves that occur because
of a disturbance in the pipe system (e.g., valve closure,
pump trip). The wave characteristics are handled using
pressure waves, which represent rapid pressure and asso-
ciated flow changes that travel at sonic velocity through
the liquid-pipe medium. A pressure wave is partially trans-
mitted and reflected at all discontinuities in the pipe sys-
tem (e.g., pipe junctions, pumps, open or closed ends,
surge tanks). The pressure wave will also be modified by
pipe wall resistance. This description is one that closely
represents the actual mechanism of transient pipe flow
(Wood et al, 2005; Boulos et al, 2004; Thorley, 1991).
Differences in the two approaches. Both the MOC and
WCM obtain solutions at intervals of t at all junctions
and components. However, the MOC also requires solu-
tions at all interior points for each time step. This require-
ment basically handles the effects of pipe wall friction
and the wave propagation characteristics of the solutions.
The WCM handles these effects by using the pressure
wave characteristics. The waves propagate through pipes
at sonic speed and are modified for the effects of friction
by a single calculation for each pipe section.
Both the Eulerian MOC and the Lagrangian WCM
will virtually always produce the same results when the
same data and model are used to the same accuracy. The
main difference is in the number of calculations; the
Lagrangian approach has an advantage.
Objectives. The primary objectives of this research
were to (1) evaluate the ability of the MOC and the
WCM to solve the basic partial differential transient pipe
flow equations for pipe systems of varying degrees of
complexity and (2) compare the solution accuracy and
computational efficiency of the two methods.
COMPUTER MODELS FOR TRANSIENT FLOW ANALYSIS
Both the MOC and WCM techniques were encoded in
the Fortran 90 programming language and implemented
for pipe system transient analysis in general-purpose com-
puter models that use the same steady-state network flow
hydraulics. This ensured that the implementation of each
solution method was as consistent as possible. For all
examples tested, the results were verified using an MOC-
based computer model (Axworthy et al, 1999) and a
WCM-based computer model (Boulos et al, 2003; Wood
& Funk 1996). The comparisons discussed here were
made using these modeling programs.
For both the MOC and WCM solutions, it is necessary
to determine a computational time interval such that the
Pump station
1
2
3
4
5
6
10
8
7
9
11
12
13
14
15
20
21
22
33
34
36
28
29
27
30
35
31
25
23
24
26
18
17
16
32
19
Overhead tank
FIGURE 4 Pipe network schematic for Example 3
Node number
pressure wave travel times will be approximately a mul-
tiple of this time and this integer multiple will be calcu-
lated for each line segment. Some adjustment is normally
required to obtain a time interval that is not unreasonably
small. This may amount to actually analyzing a system
with lengths (or wave speeds) slightly different from the
true values; a tolerance should be chosen so that this is an
acceptable deviation from the actual situation. Initial
flow conditions in all line segments
and static pressure head (or pressure)
at all junctions and components must
be known. The required initial pres-
sure head conditions may be static
heads (P/) or hydraulic grade lines
(elevation + P/).
Before transient calculations are
initiated, all components must ini-
tially be in a balanced state (i.e., the
initial pressure change across the
component and flow through the
component should be compatible
with the characteristic relationship
for that component). The initial flow
for each component must be known,
and the pressure on each side of the
component must be defined. Finally,
the exact nature of the disturbance
must be specified. The disturbance
will normally be a known change in
the stem position for a valve, a
change in the operational speed for a
pump, or the loss of power to a
pump (trip).
Results for pressure head and
flow variations are calculated for
each time step at all components and
junctions in the pipe system. For the
MOC approach, these calculations
are also required at all interior loca-
tions. For the WCM approach, a sin-
gle calculation is carried out for each
pressure wave to determine the effect
of pipe friction as the wave is trans-
mitted through the pipeline.
Numerical results. Justification for
the use of any transient flow algo-
rithm rests on its ability to solve
problems by means of computer
implementation. This is best evalu-
ated by comparing solutions ob-
tained using the various approaches.
This research compared solutions
for a number of water distribution
systems of various sizes using an
equivalent time step. The number of
calculations required to obtain the
solutions were also compared. This number was pro-
portional to the execution time needed to perform a tran-
sient analysis and therefore constituted a good indicator
of the computational efficiency of the numerical-solu-
tion procedures.
Example 1. In this example, water is flowing from a
reservoir at the upstream end to the downstream end of
a line of constant cross-sectional area A and of length L,
Times
The Lagrangian WCM and the Eulerian MOC produced virtually identical results
indicated by the single line.
P
r
e
s
s
u
r
e

H
e
a
d
f
t

(
m
)
0
50
(15.24)
100
(30.48)
150
(45.72)
200
(60.96)
250
(76.20)
300
(91.44)
350
(106.68)
400
(121.92)
0 5 10 15 20 25 30 35 40 45 50 55 60
FIGURE 5 Pressure head at node 1 for Example 3
100 (30.48)
105 (32.00)
110 (33.52)
115 (35.05)
120 (36.58)
125 (38.10)
130 (39.62)
135 (41.15)
140 (42.67)
145 (44.20)
150 (45.72)
155 (47.24)
160 (48.77)
165 (50.29)
170 (51.81)
175 (53.34)
180 (54.86)
0 5 10 15 20 25 30 35 40 45 50 55 60
Times
The Lagrangian WCM and the Eulerian MOC produced virtually identical results indicated by
the single line.
P
r
e
s
s
u
r
e

H
e
a
d
f
t

(
m
)
FIGURE 6 Pressure head at node 19 for Example 3
a general pipeline profile with initial uniform velocity,
V
0
(or flow rate, Q
0
), and a wave speed, c. At time t = 0,
a butterfly valve located at the downstream end of the line,
which is completely open, begins to close.
Figure 1 shows two instances of this example. In the
first case, there is no orifice at the reservoir entrance,
whereas in the second case an orifice is added at the
entrance. Both cases were analyzed using an exact solution
of the basic partial differential equations (Eqs 1 and 2), and
the actual results were compared with MOC and WCM
results. Analytical and numerical solution details are pro-
vided elsewhere (Boulos et al, 2004; Boulos et al, 1990).
Calculations were carried out using the data shown
in Figure 1 (cases 1 and 2) for a complete valve closure (c)
Pipe Length Diameter Node Elevation Demand
Number ft (m) in. (mm) Roughness Number ft (m) gpm (L/s)
1 2,400 (732) 12 (300) 100 1 50 (15) 694.4 (44)
2 800 (244) 12 (300) 100 2 100 (30) 8 (0.5)
3 1,300 (396) 8 (200) 100 3 60 (18) 14 (0.9)
4 1,200 (366) 8 (200) 100 4 60 (18) 8 (0.5)
5 1,000 (305) 12 (300) 100 5 100 (30) 8 (0.5)
6 1,200 (366) 12 (300) 100 6 125 (38) 5 (0.3)
7 2,700 (823) 12 (300) 100 7 160 (49) 4 (0.3)
8 1,200 (366) 12 (300) 140 8 110 (34) 9 (0.6)
9 400 (122) 12 (300) 100 9 180 (55) 14 (0.9)
10 1,000 (305) 8 (200) 140 10 130 (40) 5 (0.3)
11 700 (213) 12 (300) 100 11 185 (56) 34.78 (2.2)
12 1,900 (579) 12 (300) 100 12 210 (64) 16 (1)
13 600 (183) 12 (300) 100 13 210 (64) 2 (0.1)
14 400 (122) 12 (300) 100 14 200 (61) 2 (0.1)
15 300 (91) 12 (300) 100 15 190 (58) 2 (0.1)
16 1,500 (457) 8 (200) 100 16 150 (46) 20 (1.3)
17 1,500 (457) 8 (200) 100 17 180 (55) 20 (1.3)
18 600 (183) 8 (200) 100 18 100 (30) 20 (1.3)
19 700 (213) 12 (300) 100 19 150 (46) 5 (0.3)
20 350 (107) 12 (300) 100 20 170 (52) 19 (1.2)
21 1,400 (427) 8 (200) 100 21 150 (46) 16 (1.0)
22 1,100 (335) 12 (300) 100 22 200 (61) 10 (0.6)
23 1,300 (396) 8 (200) 100 23 230 (70) 8 (0.5)
24 1,300 (396) 8 (200) 100 24 190 (58) 11 (0.7)
25 1,300 (396) 8 (200) 100 25 230 (70) 6 (0.4)
26 600 (183) 12 (300) 100 27 130 (40) 8 (0.5)
27 250 (76) 12 (300) 100 28 110 (34) 0 (0)
28 300 (91) 12 (300) 100 29 110 (34) 7 (0.4)
29 200 (61) 12 (300) 100 30 130 (40) 3 (0.2)
30 600 (183) 12 (300) 100 31 190 (58) 17 (1.1)
31 400 (122) 8 (200) 100 32 110 (34) 17 (1.1)
32 400 (122) 8 (200) 100 33 180 (55) 1.5 (0.1)
34 700 (213) 8 (200) 100 34 190 (58) 1.5 (0.1)
35 1,000 (305) 8 (200) 100 35 110 (34) 0 (0)
36 400 (122) 8 (200) 100 36 110 (34) 1 (0.1)
37 500 (152) 8 (200) 100 26 235 (72) Tank
38 500 (152) 8 (200) 100
39 1,000 (305) 8 (200) 100
40 700 (213) 8 (200) 100
41 300 (91) 8 (200) 100
TABLE 4 Network characteristics for example 3
occurring over a time t
c
= 10 s. A com-
putational time period of 1.0 s (L/c)
is necessary. is the ratio of the effec-
tive flow area to the fully open flow
area for the butterfly valve. For all
three methods (exact, MOC, and
WCM), the computations were initi-
ated at the valve using the initial con-
ditions and the value for the valve area
ratio at the end of the first time
period. The computations then pro-
ceeded using results obtained from the
previous calculations. Results for the
three methods of analysis are shown in
Table 1 (case 1), Table 2 (case 2), and
Figure 2. Tables 1 and 2 compare val-
ues for the flow rate and the pressure
head at the valve for time intervals of
1.0 s. For this example and for both
cases analyzed, the two numerical
methods produced results that were
identical to the exact solution.
Example 2. The network used in
the second example was studied by
other researchers (Streeter & Wylie,
1967) and is shown in Figure 3, part
A. The network comprises nine pipes,
five junctions, one reservoir, three
closed loops, and one valve located at
the downstream end of the system.
The valve is shut to create the tran-
sient. Table 3 summarizes the perti-
nent pipe system characteristics; the
reservoir level of 626.64 ft (191 m) is
shown in Figure 3, part A.
Parts B and C of Figure 3 compare
the transient results obtained using the
MOC and WCM solution approaches
at node 3 and the valve, respectively. A
20-ft (67-m) length tolerance was used
in the analysis, which resulted in a
required time step of 0.1 s. In the fig-
ures, both solutions were plotted; the
two methods produced results that are
virtually indistinguishable.
Example 3. The methods were
applied to a slightly larger, more com-
plex system (Figure 4). This network
represents an actual water system and
consists of 40 pipes, 35 junctions, 1
supply pump, and 1 tank. This exam-
ple was taken from the EPANET doc-
umentation (Rossman, 1993).
Table 4 summarizes the pertinent
pipe system characteristics. The pump
station is modeled by designating the
FIGURE 7 Pipe network schematic for Example 4
797 pipes
1 pump
5 tanks
Maximum length = 4,200 ft (1,280 m), minimum length = 20 ft (6 m)
0
25 (7.62)
50 (15.24)
75 (22.86)
100 (30.48)
125 (38.10)
150 (45.72)
175 (53.34)
200 (60.96)
225 (68.58)
250 (76.20)
275 (83.82)
300 (91.44)
325 (99.06)
350 (106.68)
0 10 20 30 40 50 60 70 80 90 100 110 120
Times
The Lagrangian WCM and the Eulerian MOC produced virtually identical results, as

P
r
e
s
s
u
r
e

H
e
a
d
f
t

(
m
)
FIGURE 8 Comparison of results at pump for Example 4
inflow at that location. Figures 5
and 6 compare the transient
results obtained using the MOC
and the WCM solution schemes
at nodes 1 and 19, respectively,
following a pump shutdown
simulated by reducing the inflow
to zero over a period of 6 s. A
20-ft (6-m) length tolerance was
used in the analysis, resulting in
a required time step of 0.0139 s.
As the figures indicate, both
methods yielded virtually iden-
tical results.
Example 4. To illustrate the
comparable accuracy of both
transient solution schemes on a
larger, more complex system, the
methods were applied to the net-
work shown in Figure 7. This
network represents an actual
water distribution system con-
sisting of 797 pipes, 581 junc-
tions, 1 supply pump, and 5
tanks. (Because of the amount
of data required and the fact
that this model was based on an
actual system, the data are not
included here.)
Pipe lengths varied from 20
to 4,200 ft (6 to 1,280 m) and
diameters from 4 to 24 in. (100
to 600 mm). Figures 8 and 9
compare the transient results
obtained using MOC and WCM
solution schemes following a
pump trip. Figure 8 shows the
pressure transient just down-
stream from the pump, whereas
Figure 9 shows results at a node
some distance away from the
pump. The pump trip was mod-
eled using the four quadrant
pump characteristics in the form
developed by Marchal and co-
workers (1985). A 20-ft (6-m)
length tolerance was used in the
analysis, resulting in a time step
of 0.0056 s. As with previous
examples, the figures indicate
that both methods produced vir-
tually identical results.
DISCUSSION
Required calculations. Both
the MOC and the WCM require
200 (60.96)
205 (62.48)
210 (64.01)
215 (65.53)
220 (67.06)
225 (68.58)
230 (70.10)
235 (71.63)
240 (73.15)
245 (74.68)
250 (76.20)
255 (77.72)
260 (79.25)
265 (80.77)
270 (82.38)
275 (83.82)
280 (85.34)
0 10 20 30 40 50 60 70 80 90 100 110 120
Times
The Lagrangian WCM and the Eulerian MOC produced virtually identical results, as
P
r
e
s
s
u
r
e

H
e
a
d
f
t

(
m
)
FIGURE 9 Comparison of results at node 3066 for Example 4
Calculations/t
Time Number of
Number Number t Interior MOC/
Example* of Nodes of Pipes s Points MOC WCM WCM
Example 2 7 9 0.1 41 48 16 3.0
Example 3 36 40 .0139 680 716 76 9.4
Example 4 589 788 .0056 15,117 15,708 1,377 11.4
Example 5 1,170 1,676 .0067 81,508 82,678 2,846 29.0
Example 6 1,849 2,649 .0056 159,640 161,486 4,495 35.9
*Examples 5 and 6 are for large existing water distribution systems modeled but not described in the article.
TABLE 5 Calculation requirements for example systems
1,200
(365.76)
1,000
(304.80)
800
(243.84)
600
(182.88)
400
(121.92)
200
(76.20)
0
0 10 20 30 40 50 60
Times
P
r
e
s
s
u
r
e

H
e
a
d
f
t

(
m
)
Low friction
Normal
High friction
FIGURE 10 Effect of pipe friction on pressure transient
many calculations to solve the transient flow problem.
These calculations involve updating the pressure and
flow at required locations at increments of the time step
t. In order to compare the number of calculations
required, the authors defined one calculation as the oper-
ation required to update the pressure and flow at a sin-
gle location.
The MOC requires a calculation at all nodes and all
interior points at each time step, whereas the WCM
requires a calculation at each node and one calculation for
each pipe at each time step. The pipe calculations are
required to modify the pressure waves in that pipe to
account for the effect of pipe wall and fittings friction.
The time step used in the analysis is determined by
the tolerance set for the accuracy of the model pipe
lengths. A time step must be chosen such that pressure
waves traverse each pipe segment in a time that is a mul-
tiple of the time step. For the comparisons shown, the
length tolerance was set to 20 ft (6 m). This means that
the largest possible time increment was chosen so that
the maximum error in the length of the pipes in the model
would not exceed 20 ft (6 m).
Table 5 summarizes the calculation requirements for
the three example systems (examples 24). In addition, the
table includes data for two additional larger existing
water distribution systems (examples 5 and 6) that have
been modeled but are not described in this article.
The number of calculations for the WCM per time
step does not change with accuracy. For the MOC, the
number of calculations per time step was roughly pro-
portional to the accuracy. For the examples given, the
calculations/t required for the MOC would roughly
double if an accuracy of 10 ft (3 m) is required and would
be halved if an accuracy of 40 ft (12 m) is called for.
Handling pipe friction. The ability of the WCM to accu-
rately model pipe friction in networks using just one cal-
culation was substantiated by the virtually identical results
obtained for the examples given here. This accuracy held
true even though pipe friction has a significant effect on
the solution. Figure 10 compares the transient analysis
with and without calculating the effect of pipe wall fric-
tion for example 2. As the figure shows, the pressure
transient was significantly modified by pipe friction.
The excellent agreement between the MOC and
WCM solutions for this system (shown in Figure 3,
part C) confirms that the computed effect of wall fric-
tion is similar for the two methods. Figure 10 also illus-
trates the growing significance of increasing the pipe
resistance. Agreement for the two methods is excellent
despite some long pipes and a wide range between max-
imum and minimum pipe lengths and diameters for the
examples.
CONCLUSIONS
Both the MOC and WCM methods are capable of ac-
curately solving for transient pressures and flows in
water distribution networks including the effects of
pipe friction. The MOC requires calculations at interior
REFERENCES
Axworthy, D.A.; McInnis, D.M.; & Karney, B.W.,
1999. TransAM Technical Reference Man-
ual. Hydratek Associates, Toronto.
Boulos, P.F.; Lansey, K.E.; & Karney, B.W., 2004.
Comprehensive Water Distribution Sys-
tems Analysis Handbook for Engineers
and Planners. MWH Soft Inc., Pasadena,
Calif.
Boulos, P.F.; Wood, D.J.; & Lingireddy, S.,
2003. Users Guide for H
2
OSURGEThe
Comprehensive Transient/Surge Analy-
sis Solution for Water Distribution
Design, Operation, Management, and
Protection. MWH Soft Inc., Pasadena,
Calif.
Boulos, P.F.; Wood, D.J.; & Funk, J.E., 1990. A
Comparison of Numerical and Exact Solu-
tions For Pressure Surge Analysis. Proc.
6th Intl. BHRA Conf. on Pressure Surges.
British Hydromech. Res. Assn., Cam-
bridge, England.
Chaudhry, H.M., 1987. Applied Hydraulic Tran-
sients. Van Nostrand Reinhold Co., New
York.
Clark, R.M. et al, 1996. Mixing in Distribution
System Storage Tanks: Its Effect on Water
Quality. Jour. Envir. Engrg.ASCE,
122:9:814.
Funk, J.E. et al, 1999. Pathogen Intrusion Into
Water Distribution Systems Due to Tran-
sients. Proc. ASME/JSME Joint Fluids
Engrg. Conf., San Francisco, Calif.
Karim, M.R.; Abbaszadegan, M.; & LeCheval-
lier, M.W., 2003. Potential for Pathogen
Intrusion During Pressure Transients.
Jour. AWWA, 95:5:134.
Karney, B.W. & McInnis, D., 1990. Transient
Analysis of Water Distribution Systems.
Jour. AWWA, 82:7:62.
Kirmeyer, G.J. et al, 2001. Pathogen Intrusion
Into the Distribution System. AwwaRF and
AWWA, Denver.
Larock, B.E.; Jeppson, R.W.; & Watters, G.Z.,
1999. Hydraulics of Pipeline Systems. CRC
Press, Boca Raton, Fla.
LeChevallier, M.W. et al, 2003. Potential for
Health Risks From Intrusion of Contami-
nants Into the Distribution System From
For the same modeling accuracy, the wave characteristic
method will normally require fewer calculations and provide
faster execution times.
points to handle the wave propagation and the effects
of pipe friction. The WCM handles these effects using
pressure waves. Therefore, for the same modeling accu-
racy, the WCM will usually require fewer calculations
and provide faster execution times. In addition, the
number of calculations per time step does not increase
for the WCM when greater accuracy is required. How-
ever, because of the difference in calculation require-
ments and the comparable accuracy of the two tech-
niques, use of the WCM may be more suitable for
analyzing large pipe networks.
The results given here compared favorably with results
obtained by other researchers (Rossman & Boulos, 1996)
for solving the transient water quality transport equa-
tions. In their study, Rossman and Boulos made detailed
comparisons of accuracy, computation time, and com-
putation storage requirements for the Eulerian and
Lagrangian numerical solution schemes. They found that
all methods produced virtually identical results, but the
Lagrangian approach was more versatile and more time-
and memory-efficient than the Eulerian approach when
modeling chemical constituents.
Any transient analysis is subject to inaccuracies because
of incomplete information regarding the piping system,
its components and degree of skeletonization, and some
uncertainty with respect to initial flow distribution (Mar-
tin, 2000). However, the efficacy of transient modeling
is enhanced by ensuring proper network model con-
struction and calibration. Properly developed and cali-
brated models for transient analysis greatly improve the
ability of water utilities to determine adequate surge pro-
tection, strengthen the integrity of their systems, and
forge closer ties with their customers as well as the sur-
rounding community. Water utility engineers can effec-
tively use these models to predict unacceptable operating
conditions developing in their distribution systems, iden-
tify risks, formulate and evaluate sound protective mea-
sures, and implement improved operational plans and
security upgrades.
ABOUT THE AUTHORS
Don J. Wood is professor emeritus in
the Department of Civil Engineering
at the University of Kentucky in Lex-
ington. A winner of the ASCE Huber
Research Prize and the 2004 Simon
Freese Environmental Engineering
Award, he has more than 40 years of
experience in the area of steady-state
and transient modeling. He has pub-
lished more than 90 technical papers and is the author
of the textbook Pressure Wave Analysis of Transient
Flow in Pipe Distribution Systems. Wood earned his
bachelors, masters, and doctoral degrees from
Carnegie Mellon University in Pittsburgh, Pa. Srinivasa
Lingireddy is an associate professor in the University of
Kentucky Department of Civil Engineering. Paul F.
Boulos (to whom correspondence should be addressed)
is president and COO of MWH Soft Inc., 370 Inter-
locken Blvd., Suite 300, Broomfield, CO 80021; e-mail
[email protected]. Bryan W. Karney is a pro-
fessor in the Department of Civil Engineering at the
University of Toronto in Ontario. David L. McPherson
is a supervising engineer with MWH Americas Inc. in
Cleveland, Ohio.
Pressure Transients. Jour. Water &
Health, 1:1.
LeChevallier, M.W., 1999. The Case for Main-
taining a Disinfectant Residual. Jour.
AWWA, 91:1:86.
Marchal, M.; Flesh, G.; & Suter, P., 1985. Calcu-
lations of Water Hammer Problems by
Means of the Digital Computer. ASME
Intl. Sym. on Water Hammer in Pumped
Storage Projects, Chicago.
Martin, C.S., 2000. Hydraulic Transient Design
for Pipeline Systems. Water Distribution
Systems Handbook (Larry W. Mays, edi-
tor). McGraw-Hill, New York.
Rossman, L.A. & Boulos, P.F., 1996. Numerical
Methods for Modeling Water Quality in
Distribution Systems: A Comparison. Jour.
Water Resources Planning & Mngmt.
ASCE, 122:2:137.
Rossman, L.A., 1993. Users Manual for
EPANET. US Environmental Protection
Agency, Drinking Water Res. Div.,
Risk Reduction Engrg. Lab.,
Cincinnati.
Streeter, V.L. & Wylie, E.B., 1967. Hydraulic
Transients. McGraw-Hill, New York.
Thorley, A.R.D., 1991. Fluid Transients in
Pipeline Systems. D. & L. George Ltd.
Publ., Herts, England.
USACE (US Army Corps of Engineers), 1999.
Liquid Process Piping Engineering Man-
ual. EM 1110-1-4008, Washington.
Watters, G.Z., 1984 (2nd ed.). Analysis and
Control of Unsteady Flow in Pipelines.
Ann Arbor Science, Boston.
Wood, D.J.; Lingireddy, S.; & Boulos, P.F., 2005.
Pressure Wave Analysis of Transient
Flow in Pipe Distribution System. MWH
Soft Inc., Pasadena, Calif.
Wood, D.J. & Funk, J.E., 1996. Surge Reference
ManualComputer Analysis of Transient
Flow in Pipe Networks. University of Ken-
tucky, Civil Engrg. Software Ctr., Lexing-
ton, Ky.
Wood, D.J.; Dorsch, R.G.; & Lightner, C., 1966.
Wave Plan Analysis of Unsteady Flow in
Closed Conduits. Jour. Hydraulics Div.
ASCE, 92:2:83.
If you have a comment about this article,
please contact us at [email protected].

Numerical Methods For Modeling Transient Flow: in Distribution Systems

Uploaded by

Copyright:

Available Formats

Numerical Methods For Modeling Transient Flow: in Distribution Systems

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Numerical Methods For Modeling Transient Flow: in Distribution Systems

Uploaded by

Copyright:

Available Formats

104 JULY 2005 | JOURNAL AWWA 97: 7 | PEER-REVI EWED | WOOD ET AL

2005 American Water Works Association

You might also like